NFS: Fix fdatasync/fsync() when confronted with a server reboot

If the server reboots before it can commit the unstable writes to disk,
then nfs_commit_release_pages() will detect this when it compares the
verifier returned by COMMIT to the one returned by WRITE. When this
happens, the client needs to resend those writes in order to guarantee
that they make it to stable storage.

This patch adds a signalling mechanism to notify fsync() that it
needs to retry all writes before it can exit.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>

3 files changed, 35 insertions, 22 deletions

diff --git a/fs/nfs/file.c b/fs/nfs/file.c
index 6a7fcab7ecb3..cc9b56691bef 100644
--- a/fs/nfs/file.c
+++ b/fs/nfs/file.c
@@ -259,7 +259,7 @@ nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
 	struct dentry *dentry = file->f_path.dentry;
 	struct nfs_open_context *ctx = nfs_file_open_context(file);
 	struct inode *inode = dentry->d_inode;
-	int have_error, status;
+	int have_error, do_resend, status;
 	int ret = 0;
 
 	dprintk("NFS: fsync file(%s/%s) datasync %d\n",
@@ -267,15 +267,23 @@ nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
 			datasync);
 
 	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
+	do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
 	have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
 	status = nfs_commit_inode(inode, FLUSH_SYNC);
-	if (status >= 0 && ret < 0)
-		status = ret;
 	have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
-	if (have_error)
+	if (have_error) {
 		ret = xchg(&ctx->error, 0);
-	if (!ret && status < 0)
+		if (ret)
+			goto out;
+	}
+	if (status < 0) {
 		ret = status;
+		goto out;
+	}
+	do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
+	if (do_resend)
+		ret = -EAGAIN;
+out:
 	return ret;
 }
 EXPORT_SYMBOL_GPL(nfs_file_fsync_commit);
@@ -286,13 +294,15 @@ nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 	int ret;
 	struct inode *inode = file->f_path.dentry->d_inode;
 
-	ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
-	if (ret != 0)
-		goto out;
-	mutex_lock(&inode->i_mutex);
-	ret = nfs_file_fsync_commit(file, start, end, datasync);
-	mutex_unlock(&inode->i_mutex);
-out:
+	do {
+		ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
+		if (ret != 0)
+			break;
+		mutex_lock(&inode->i_mutex);
+		ret = nfs_file_fsync_commit(file, start, end, datasync);
+		mutex_unlock(&inode->i_mutex);
+	} while (ret == -EAGAIN);
+
 	return ret;
 }
 
diff --git a/fs/nfs/nfs4file.c b/fs/nfs/nfs4file.c
index eb5eb8eef4d3..eef1b38a1b08 100644
--- a/fs/nfs/nfs4file.c
+++ b/fs/nfs/nfs4file.c
@@ -95,16 +95,18 @@ nfs4_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 	int ret;
 	struct inode *inode = file->f_path.dentry->d_inode;
 
-	ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
-	if (ret != 0)
-		goto out;
-	mutex_lock(&inode->i_mutex);
-	ret = nfs_file_fsync_commit(file, start, end, datasync);
-	if (!ret && !datasync)
-		/* application has asked for meta-data sync */
-		ret = pnfs_layoutcommit_inode(inode, true);
-	mutex_unlock(&inode->i_mutex);
-out:
+	do {
+		ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
+		if (ret != 0)
+			break;
+		mutex_lock(&inode->i_mutex);
+		ret = nfs_file_fsync_commit(file, start, end, datasync);
+		if (!ret && !datasync)
+			/* application has asked for meta-data sync */
+			ret = pnfs_layoutcommit_inode(inode, true);
+		mutex_unlock(&inode->i_mutex);
+	} while (ret == -EAGAIN);
+
 	return ret;
 }
 
diff --git a/fs/nfs/write.c b/fs/nfs/write.c
index e1b5fe4d873a..9347ab7c9574 100644
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -1580,6 +1580,7 @@ static void nfs_commit_release_pages(struct nfs_commit_data *data)
 		/* We have a mismatch. Write the page again */
 		dprintk(" mismatch\n");
 		nfs_mark_request_dirty(req);
+		set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
 	next:
 		nfs_unlock_and_release_request(req);
 	}